PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Phys Med Biol. Author manuscript; available in PMC 2010 May 19.
Published in final edited form as:
PMCID: PMC2872939
NIHMSID: NIHMS201911

The impact of calibration phantom errors on dual-energy digital mammography

Abstract

Microcalcification is one of the earliest and main indicators of breast cancer. Because dual-energy digital mammography could suppress the contrast between the adipose and glandular tissues of the breast, it is considered a promising technique that will improve the detection of microcalcification. In dual-energy digital mammography, the imaged object is a human breast, while in calibration measurements only the phantoms of breast tissue equivalent materials are available. Consequently, the differences between phantoms and breast tissues will lead to calibration phantom errors. Based on the dual-energy imaging model, formulae of calibration phantom errors are derived in this paper. Then, this type of error is quantitatively analyzed using publicly available data and compared with other types of error. The results demonstrate that the calibration phantom error is large and dominant in dual-energy mammography, seriously decreasing calculation precision. Further investigations on the physical meaning of calibration phantom error reveal that the imaged objects with the same glandular ratio have identical calibration phantom error. Finally, an error correction method is proposed based on our findings.

1. Introduction

Breast cancer is a deadly disease that adversely affects the lives of many people, primarily women. Microcalcification is one of the earliest and main indicators of breast cancer, and mammography is the gold standard for breast cancer screening. Thus, the visualization and detection of microcalcifications in mammography play a crucial role in reducing the rate of mortality from breast cancer. Microcalcifications are usually smaller than 1.0 mm and mainly composed of calcium compounds (Fandos-Morera et al 1988) such as apatite, calcium oxalate and calcium carbonate. Microcalcifications have greater x-ray attenuation coefficients than the surrounding breast tissues, so they are more visible on homogeneous soft-tissue backgrounds. However, the visualization of microcalcifications could be obscured in mammograms because of overlapping of tissue structures. Tissue structures in mammograms arise from the differences of the x-ray attenuation coefficients between adipose tissue, glandular tissue, ducts, vessels and soft-tissue masses. Microcalcifications, especially smaller ones, are extremely difficult to discriminate even if the signal-to-noise ratio is high (Brettle and Cowen 1994). Therefore, microcalcification detection in mammography suffers from a high false-negative rate (Soo et al 2005).

Dual-energy digital mammography is considered as a prospective technique to improve the detection of microcalcification. The healthy breast is mainly composed of adipose, epithelial and stroma tissues which can be grouped into two attenuating types: adipose tissue and glandular tissue. In dual-energy digital mammography, high- and low-energy images of the breast are acquired using two different x-ray spectra. By exploiting the difference in x-ray attenuation between different materials at different x-ray energies, the high- and low-energy images can be synthesized to suppress the contrast between adipose and glandular tissues and subsequently generate the dual-energy calcification image. Under ideal imaging conditions, when the imaging data are free of scatter and other biases, dual-energy digital mammography could be used to determine the thickness of microcalcification and the breast glandular ratio (Lemacks et al 2002). This quantitative information can be incorporated into computer-aided detection algorithms to enhance mammographic interpretation.

However, a lot of research must be carried out prior to applying dual-energy digital mammography in clinics. Early research on the feasibility of dual-energy mammography includes optimal high- and low-energy spectra selection (Johns and Yaffe 1985), selection of inverse-map functions (Cardinal and Fenster 1990), influence of detectors (Chakraborty and Barnes 1989, Boone et al 1990, Brettle and Cowen 1994) and experimental studies (Johns et al 1985, Brettle and Cowen 1994). Lemacks et al (2002) presented a theoretical framework to calculate the (quantum) noise in dual-energy calcification image and made numerical analysis as well. Their results were presented in terms of contrast-to-noise ratio under various imaging conditions, including the x-ray spectra, microcalcification size, tissue composition and breast thickness. Based on the work of Lemacks et al (2002), Kappadath et al made a series of investigations on dual-energy digital mammography (Kappadath and Shaw 2003, 2004, 2005, Kappadath et al 2004, 2005). Kappadath and Shaw (2003) investigated various inverse-map functions for dual-energy digital mammography, and concluded that the mean fitting error is ~50 μm and the max fitting error is ~150 μm for the microcalcification thickness when using the cubic or conic functions. Kappadath and Shaw (2005) developed a scatter and nonuniformity correction technique for dual-energy digital mammography. In their implementation on the clinical equipment, microcalcifications in 300–355 μm range were clearly visible in dual-energy calcification images. The microcalcification size threshold decreased to 250–280 μm when the visible criteria were lowered to barely eyesight. Dual-energy digital mammography has also been investigated and evaluated by other investigators (Bliznakova et al 2006, Brandan and Ramirez-R 2006, Taibi et al 2003).

In dual-energy digital mammography, there are two necessary steps: the first step is calibration, which estimates the polynomial coefficients using calibration data, while the second step is the calculation of the component information of the imaged object. When dual-energy digital mammography is used clinically, the imaged object is a human breast. In calibration, however, only phantoms of a breast tissue equivalent material can be used. The composition and density differences between the phantoms and breast tissues cause differences in linear attenuation coefficients, which consequently lead to so-called calibration phantom errors. The error was first indicated and investigated by Mou and Chen (2007) and consisted of two parts: microcalcification thickness error and glandular ratio error. The small differences in linear attenuation coefficients between calibration phantoms and breast tissues lead to huge calibration phantom errors in the final results. The phantom materials, like PMMA, polyethylene and RMI phantoms, are not suitable for calibration of dual-energy digital mammography in clinics. Comparably, CIRS phantoms are a better choice.

In reality, the elemental composition ratio of human breast varies greatly (Hammerstein et al 1979). The differences in linear attenuation coefficients between calibration phantoms and breast tissues are inevitable. Therefore, calibration phantom errors cannot be removed solely by redesigning calibration phantoms. This study systematically analyzes the calibration phantom errors, investigates their physical meaning and develops possible error correction methods.

2. Theory

This section first presents the imaging principles and the physical model of dual-energy digital mammography and then develops the formulae for calibration phantom errors.

2.1. Imaging principle

The principles of dual-energy imaging have been discussed extensively in the literature (Alvarez and Macovski 1976, Lehmann et al 1981, Brody et al 1981, Johns and Yaffe 1985, Gingold and Hasegawa 1992). In the medical diagnostic x-ray imaging modality, x-ray attenuation includes photoemission and Compton scattering. The linear attenuation coefficient μ(E) is (Alvarez and Macovski 1976)

equation M1
(1)

where fKN(E) is the Klein–Nishina function,

equation M2
(2)

and α = E/510.975. The attenuation coefficient of any material above the k-edge can be approximated by equation (1). Because the atomic numbers of the elements in human body are small and their k-edge energies are low, equation (1) is valid for medical diagnostic x-ray imaging. The term E−3 approximates the energy dependence of the photoelectric interaction, and fKN(E) gives the energy dependence of the total cross-section for Compton scattering. The dependences of bpu and bdu on physical parameters are given below:

equation M3
(3)

where K1 and K2 are constants, ρ is the mass density, n ≈ 4, A is the atomic weight and Z is the atomic number. If the imaged object is composed of two known materials, A with thickness LA and B with thickness LB, the full object thickness is L = LA + LB and the attenuation M (E) of the imaged object can be expressed as (Lehmann et al 1981)

equation M4
(4)

where

equation M5
(5)

and bpuA, bduA, bpuB and bduB are known quantities of the materials A and B.

An object can be imaged under high- and low-energy spectra:

equation M6
(6)

where P0h(E) and P0l(E) represent the incident x-ray photon fluence of high- and low-energy spectra, Q(E) is the detector response, and fh and fl are the negative logarithm transmitted intensities of high- and low-energy spectra. bp and bd are defined in equation (5). If P0h(E), P0l(E) and Q(E) are known, bp and bd of the object can be solved from equation (6), then LA and LB can be obtained based on equation (5). Since P0h(E) and P0l(E) are not easy to measure, in practice researchers often use inverse-map functions to solve LA and LB (Brody et al 1981):

equation M7
(7)

The coefficients kAi and kBi, i = 0, …,7, are determined using a calibration procedure (Lehmann et al 1981). By measuring the transmission intensities of various combinations of basic materials A and B with high- and low-energy spectra, the groups of calibration data (fh, fl, LA, LB) can be acquired to estimate the coefficients kAi and kBi. Accordingly, other forms of inverse-map functions, for example, implicit polynomials (Alvarez and Macovski 1976) and rational polynomials (Cardinal and Fenster 1990) can also be used for dual-energy calculations.

2.2. Physical model

Lemacks et al (2002) proposed a numerical framework to perform the dual-energy digital mammography calculation. They assumed that there are three attenuating materials in the breast: adipose tissue (thickness ta), glandular tissue (thickness tg) and microcalcification (thickness tc). Depending on the distribution of materials, the transmitted fluence incident on the detector is given by

equation M8
(8)

where P0(E) and P(E) are the incident photon fluence on the surface of the breast and the transmitted fluence, respectively; μa(E), μg(E) and μc(E) are the linear attenuation coefficients of adipose tissue, glandular tissue and microcalcification, respectively.

Because of the characteristics of diagnostic x-ray imaging physics shown in equation (1), only two unknowns can be solved for dual-energy digital mammography. During mammography, the breast is usually compressed to a uniform thickness T that can be easily measured. The contribution of microcalcifications to the total breast thickness can be ignored because the microcalcifications are small in size and sparsely present, i.e. Tta + tg. With the total breast thickness T known, the three unknowns ta, tg and tc can be expressed as two unknowns: glandular ratio g = tg/Ttg/(ta + tg) and microcalcification thickness tc (Kappadath and Shaw 2003). Now, equation (8) can be rewritten as

equation M9
(9)

In dual-energy imaging calculations, a reference signal Ir is needed in order to increase the dynamic range of the intensity values. The reference signal can be measured by a certain breast phantom. The exposure data f is defined as the logarithmic value of the ratio of transmitted exposure I to reference signal Ir. The high- and low-energy logarithmic intensities fh(tc, g) and fl (tc, g), respectively, are measured independently using x-ray beams at different kVps:

equation M10
(10)

In this paper, we assume that there is no scatter because it can be well corrected (Kappadath and Shaw 2005). Therefore, g and tc can be solved by a dual-energy imaging method.

2.3. Calibration phantom error formulae

In dual-energy digital mammography, phantoms for calibration are always different from real breast tissues that will be imaged. These differences lead to calibration phantom errors Δtc of the microcalcification thickness and Δg of the glandular ratio.

When phantoms are used for calibration, equation (10) can be expressed as

equation M11
(11)

where the second row of equation (11) represents the imaging of breast tissues and the third row represents the imaging of phantoms. The superscript ‘hu’ indicates breast tissues while ‘ph’ indicates phantoms. g and tc represent the true glandular ratio and microcalcification thickness of the breast, respectively. Let g′ and equation M12 represent the calculated glandular ratio and microcalcification thickness of the phantoms:

equation M13
(12)

equation M14
(13)

where Δg and Δtc are calibration phantom errors. The third row of equation (11) can be approximated by the first-order Taylor series expansion of variables g and tc, so that Δg and Δtc can be separated:

equation M15
(14)

where equation M16 and equation M17 are defined as

equation M18
(15)

and

equation M19
(16)

Let equation M20. By substituting the attenuation coefficients of phantoms for those of human breast on the left side of equation (14), expanding the left side, and eliminating the same terms on both sides, we have

equation M21
(17)

Define the symbols

equation M22
(18)

and there is

equation M23
(19)

where equation M24 and equation M25 are the average attenuation coefficients of microcalcification calibration phantom weighted by transmitted high- and low-spectra. equation M26 and equation M27 are the average attenuation coefficient differences between the glandular calibration phantom and the adipose calibration phantom weighted by transmitted high- and low-spectra. equation M28 (m = c, g, a; j = h, l) is the average attenuation coefficient difference between calibration phantoms and human breast for microcalcifications, glandular and adipose weighted by transmitted high- and low-spectra. In this paper, we focus on the calibration phantom errors caused by breast calibration phantoms and assume that equation M29 and equation M30 are zero. From equation (19), the calibration phantom errors Δtc and Δg can be determined as

equation M31
(20)
equation M32
(21)

3. Data for calculation

Using the formulae derived in section 2, the calibration phantom errors can be calculated using publicly available data. The imaging conditions are similar to those in Kappadath and Shaw (2003, 2005) and Lemacks et al (2002), which agree with the clinical mammography system. The x-ray spectra are 25 kVp and 50 kVp with a Mo anode and a 0.03 mm Mo filter. The spectra data are obtained from the classical handbook (Fewell and Shuping 1978). The detector consists of a CsI:Tl converter layer coupled with an aSi:H + TFT flat-panel detector. In our calculation, the scintillator thickness ts is 45 mg cm−2 for CsI:Tl. All photons transmitted through the imaged object are assumed to be absorbed completely in the perfectly efficient converter layer (Lemacks et al 2002). The detector response function Q(E) is defined as

equation M33
(22)

equation M34
(23)

where μs(E) is the mass attenuation coefficient of CsI (Berger et al 2005).

Hammerstein et al (1979) have determined the elemental compositions of glandular and adipose tissues for human breast. The composition per weight of H, N and P was well determined. However, the composition per weight of C and O presents a wide range of possible values. Table 1 lists two extreme compositions 1 and 2 and one general composition 3. In this paper, the composition per weight of C and O refers to ‘composition 3’ without specific notification. The densities equal to 0.93 g cm−3 for adipose tissue and 1.04 g cm−3 for glandular tissue.

Table 1
Elemental compositions (percentage per weight) of C and O for glandular and adipose tissues (Hammerstein et al 1979).

Phantom materials are used for calibration. While polyethylene (CH2) is commonly used for an adipose phantom material, PMMA (C5H8O2) and acrylic (C3H4O2) are often used for a glandular phantom material. Also, commercial breast tissue equivalent materials are available in the market. For example, breast adipose and glandular tissue equivalent materials from CIRS (Computerized Imaging Reference Systems, Inc., Norfolk, VA, USA) are based on Hammerstein et al (1979), and those from RMI (Radiation Measurements Inc., Madison, WI, USA) are based on White et al (1977). Our previous work (Mou and Chen 2007) showed that CIRS phantoms used in calibration lead to smaller calibration phantom errors compared with other phantoms. Therefore, we use CIRS phantoms in this paper. The elemental compositions and densities of CIRS phantoms and human breast are listed in table 2 (Byng et al 1998).

Table 2
Elemental compositions (percentage per weight) and densities of CIRS phantoms and breast tissues (Hammerstein et al 1979, Byng et al 1998).

Fandos-Morera et al (1988) analyzed the compositions of microcalcifications and setup possible relationships with malignant and benign lesions. Calcium oxalate (CaC2O4) is a common component of microcalcification in human breasts. In our study, calcium oxalate is employed as microcalcifications with a density of 2.20 g cm−3. Based on the elemental compositions listed in table 2, mass attenuation coefficients of CIRS phantoms and breast tissues are calculated via the database of XCOM from NIST (Berger et al 2005). The breast thickness is assumed to be 5 cm.

4. Results

4.1. Calibration phantom error

Microcalcifications are relatively small in size, ranging from 1 mm to less than 100 μm, with most of them smaller than 500 μm. Therefore, we will focus on microcalcifications smaller than 500 μm with glandular ratio ranging from 0% to 100%.

CIRS phantoms are adopted for calibration. First, we calculate the average difference of linear attenuation coefficients between CIRS phantoms and breast tissues. According to the elemental compositions listed in table 2, we can obtain the mass attenuation coefficients using the XCOM database, and then calculate the average difference as

equation M35
(24)

where μ(E)/ρ is the mass attenuation coefficient and ρ is the mass density. In the energy range 10–50 keV, the average difference between CIRS phantoms and breast tissues is very small: 1.14% for adipose and 0.95% for glandular. Hence, the attenuation characteristics of CIRS phantoms can mimic those of breast tissues well.

Calibration phantom errors are calculated using equations (20) and (21) with the spectra data, detector response function, mass attenuation coefficients and densities of phantoms and breast tissues mentioned in section 3. Tables 3 and and44 list the calculated calibration phantom errors for different combinations of the microcalcification thickness and glandular ratio for composition 3, showing that the calibration phantom errors are large enough to seriously deteriorate the calculation precision.

Table 3
Calibration phantom errors for the microcalcification thickness Δtc (μm).
Table 4
Calibration phantom errors for glandular ratio Δg (%).

4.2. Calibration phantom error when breast composition varying

The breast compositions of different individuals vary greatly. When calibration phantoms are fixed, the variations of different individuals result in different calibration phantom errors. Tables 5 and and66 list the calibration phantom errors Δtc and Δg for the three compositions listed in table 1 when the microcalcification thickness is 250 μm and the glandular ratio ranges from 0% to 100%. In tables 5 and and6,6, calibration phantom errors have significant differences for different compositions. For example, when the glandular ratio is 50%, the error Δtc varies from −270 μm to 628 μm as the breast composition changes from composition 1 to composition 2. Therefore, the true microcalcification thickness cannot be determined if the breast composition is unknown. In other words, the computed pair ( equation M36, g′) will map to different combinations (tc, g) of the microcalcification thickness and glandular ratio for different breast compositions. Figure 1 illustrates how the computed pair ( equation M37, g′) maps to a true pair (tc, g) when breast composition varies. Two cases are demonstrated, which are (850 μm, 35%) and (750 μm, 60%) of the computed pair ( equation M38, g′). As demonstrated in figure 1, for the computed pair ( equation M39, g′), the corresponding true value (tc, g) changes along a line segment connecting the two extreme cases of compositions 1 and 2. Obviously, it is impossible for the mapped real pair (−130 μm, 114%) of composition 2 to have the case (750 μm, 60%). In practice, the computed pair cannot be (750 μm, 60%) when the breast tissue is of composition 2. Here, we just plot it in the figure for clarity. In practice, the breast composition will change between the two extreme compositions 1 and 2 in the shadow.

Figure 1
Maps of computed pair ( equation M62, g′) to real pair (tc, g) when breast composition changes.
Table 5
Calibration phantom errors for the microcalcification thickness Δtc (μm).
Table 6
Calibration phantom errors for glandular ratio Δg (%).

5. Discussion

5.1. Magnitude of the calibration phantom error

The data calculation results in section 4 show that the calibration phantom errors are large enough that they greatly decrease the calculation precision of dual-energy digital mammography. Although the used CIRS phantoms have only about a 1% difference in linear attenuation coefficients compared to the general breast tissue composition, the calibration phantom errors could reach tens to hundreds of percents. For example, when a microcalcification is 250 μm, the calibration phantom error Δtc ranges from −27% (−67 μm) to 170% (425 μm) as the glandular ratio increases from 0% to 100%. Calibration phantom errors also change significantly as the composition varies. For an imaged object of 250 μm microcalcification and 50% glandular ratio, the calibration phantom error Δtc is −270 μm for composition 1 and 628 μm for composition 2.

Because our imaging conditions are similar to those of existing references (Kappadath and Shaw 2003, 2005, Lemacks et al 2002), we can compare our results to the reported results in these papers. Lemacks et al (2002) investigated the errors caused by quantum noise of the detector with normal exposure dose. They showed that a 250 μm microcalcification could be subject to random noise of 80 μm for a 5 cm breast with 50% glandular ratio. However, our study shows that the corresponding calibration phantom error of the microcalcification thickness is 187 μm for composition 3 and even bigger for compositions 1 (−270 μm) and 2 (628 μm). Thus, the calibration phantom errors are much greater than errors caused by detector quantum noise. Kappadath and Shaw (2003) concluded that the mean fitting error is ~50 μm and the max fitting error is ~150 μm for the microcalcification thickness when adopting a cubic or conic function. In our study, when the breast is of composition 3, the mean calibration phantom error for the microcalcification thickness is 192 μm and the maximum is 433 μm (see table 3). When the breast is of composition 2, the maximum calibration phantom error will reach 813 μm (see table 5). Obviously, calibration phantom errors are much greater than the fitting errors in dual-energy digital mammography. Kappadath and Shaw (2005) showed that microcalcifications of 300–355 μm can be clearly seen while microcalcifications of 250–280 μm are barely visible in dual-energy calcification images after scatter and nonuniformity corrections. Due to the existence of large calibration phantom errors and the diversity of breast compositions, the microcalcification thickness threshold cannot be determined even if both scatter and nonuniformity corrections are well addressed. In summary, a calibration phantom error is dominant among all kinds of errors in dual-energy digital mammography calculation.

5.2. Property of the calibration phantom error

From tables 3 and and4,4, we can see that for a fixed glandular ratio, calibration phantom errors are almost unchanged when the microcalcification thickness changes from 0 μm to 500 μm. For example, Δtc changes from 193 μm to 182 μm and Δg changes from −16.2% to −15.4% when the microcalcification thickness ranges from 0 μm to 500 μm with the glandular ratio being 50%. The variation ratios are 5.7% for Δtc and 4.9% for Δg. In fact, Δtc and Δg are calculated from equations (20) and (21) which are formed by equation M40 (m = c, g, α; j = h, l), equation M41 (m = c, g; j = h, l), g and T, where equation M42 and equation M43 are the attenuation coefficients and the corresponding differences weighted by the transmitted high- and low-spectra. The transmitted high- and low-spectra are slightly affected by the microcalcification thickness because of their small size, and hence equation M44 and equation M45 change little when the microcalcification thickness changes. Therefore, for a fixed glandular ratio g, the calibration phantom errors Δtc and Δg are almost unchanged when the microcalcification thickness changes from 0 μm to 500 μm.

5.3. Analysis of physical meaning for the calibration phantom error

In section 2.1, we showed that the attenuation of an imaged object can be described by equation (4), and the pair (bd, bp) can be used to represent the attenuation of the object. In mammography, the imaged object is breast tissues plus microcalcifications. We have

equation M46
(25)

where (bd_breast, bp_breast) and (bd_mc, bp_mc) represent the attenuation of the breast tissues and microcalcifications, respectively. As shown in figure 2, we use the vectors OA, OB and BA to represent (bd, bp), (bd_breast, bp_breast) and (bd_mc, bp_mc) respectively.

Figure 2
Sketch of the attenuation characteristics in mammography.

We define a quantity of attenuation gradient (AG) as follows:

equation M47
(26)

In our study, the component of microcalcification is calcium oxalate (CaC2O4). The AG for microcalcification can be written as

equation M48
(27)

According to equation (1), bdu_CaC2 O4 and bpu_CaC2 O4 only relate to the linear attenuation coefficient of calcium oxalate. Therefore, AGmc is a fixed value in our study.

Now, we adopt a 250 μm microcalcification with 50% glandular ratio of composition 3 as the imaged object. Recall that table 3 shows that the calibration phantom error Δtc is 187 μm. Figure 3 analyzes Δtc. Because the breast thickness T is known and fixed and there exist attenuation differences between the breast tissues and the calibration phantoms, the pair (bd, bp) of the breast tissues and calibration phantoms will change along two independent line segments. The pair (bd, bp) of the breast tissues changes along the blue dotted line segment CD when glandular ratio varies from 0% to 100%. The point C corresponds to 0% glandular ratio, while point D corresponds to 100% glandular ratio. Correspondingly, the pair (bd, bp) of CIRS phantoms changes along the mauve dotted line segment CcDc. However, the real pair (bd, bp) of this imaged object is point A. From point A, we can draw a line with gradient AGmc to represent the attenuation of microcalcification. This line crosses CD and CcDc, generating two intersection points B and Bc. Note that point B maps to the attenuation of breast tissues of glandular ratio 50%, and the vector BA represents the attenuation of real microcalcification. Because point Bc maps to the attenuation of CIRS phantoms with calculated glandular ratio g′ (34.2%), the vector BCA represents the attenuation of the calculated microcalcification with thickness equation M49 (437 μm). The vector BCB represents the calibration phantom error Δtc (187 μm), which is caused by the differences between breast tissues and calibration phantom. Because the size of the microcalcification is relatively small, the attenuation of the microcalcification is much smaller compared with that of breast tissues or phantoms, and the point A is always close to the blue dotted line segment. Therefore, the length of BCB is relatively long compared with that of BA, which implies big values of the calibration phantom error Δtc.

Figure 3
Physical meaning of calibration phantom errors.

As can be seen in tables 3 and and4,4, the calibration phantom errors are almost unchanged when the glandular ratio is fixed and the microcalcification thickness varies. This phenomenon can be explained by figure 4. The blue and mauve dotted line segments, point B and point Bc, have the same meaning as in figure 3. For a glandular ratio of 50%, point B is fixed on the blue dotted line segment which maps to the attenuation of breast tissues. We can draw a red line that crosses point B with gradient AGmc. The vector BCB represents the calibration phantom error Δtc when the true glandular ratio is 50%. The three points A1, A2 and A3 correspond to attenuations of three imaged objects with 50% glandular ratio and microcalcification of 150 μm, 250 μm and 350 μm, respectively. All of them are located on the red line that crosses B and Bc, which implies the same calibration phantom errors. Therefore, there will be the same calibration phantom errors if the imaged objects have the same glandular ratio.

Figure 4
Calibration phantom errors of different microcalcification thicknesses.

However, there still exist small differences among the calibration phantom errors that can be seen in tables 3 and and44 when the glandular ratio is constant. These differences come from the dependence of energy during the calculation of bp and bd. From equations (3)(5), we can see that bp and bd are almost independent of energy. If bp and bd were independent of energy completely, there would be no difference in the calibration phantom errors. Actually, bp and bd are weakly dependent on energy (Gingold and Hasegawa 1992), and both bp and bd are influenced by the transmitted spectra. When the microcalcification thickness changes, the transmitted high- and low-energy spectra will also change slightly. Consequently, bp and bd will change a little; hence, there are small differences in calibration phantom errors.

5.4. Error correction method

Omitting the small difference caused by energy dependence, the calibration phantom errors will be constant if the imaged objects have the same glandular ratio. This property suggests a correction method for the calibration phantom error Δtc.

In dual-energy digital mammography, for an imaged breast, we can calculate (g′, equation M50) of each pixel pair based on the high- and low-energy image and set up a coordinate system with the horizontal axis as g′ and vertical axis as equation M51. (g′, equation M52) of each pixel pair can be positioned in the coordinate system. Obviously, the computed equation M53 of the non-microcalcification pixel pair equals their Δtc.

In full-field digital mammography, there are several millions of pixels in an image. Because microcalcifications are small and sparsely present in breast tissues, most pixels in mammogram are non-microcalcification pixels. With millions of pairs in the coordinate system, a curve can be fitted. (g′, equation M54) on the fitted curve can be considered as those correspond to non-microcalcification pixel pairs. Therefore, equation M55 of (g′, equation M56) on the fitted curve is the calibration phantom error Δtc.

Recall that the calibration phantom errors Δtc and Δg are constant if the imaged objects have the same glandular ratio. Denoted by ( equation M57) the calculated results for any pixel pair Pi in dual-energy mammograms, we can find the corresponding pair ( equation M58, tci0′) on the fitted curve. Therefore, tci 0′ is the calibration phantom error Δtci of the pixel pair Pi. The true microcalcification thickness of the pixel pair Pi corresponds to tci = tci′ − tci 0′. For example, the fitted red curve tc0′ = 747.4*g′ − 63.92 in figure 5 corresponds to the data in tables 3 and and4.4. For pixel pair Pi, the true microcalcification thickness and glandular ratio are 250 μm and 50%, respectively. The calculated pair ( equation M59) of Pi is (34.2%, 437 μm), as shown in figure 5. Substituting equation M60 (34.2%) into the expression of the fitted curve, tci 0′ (192 μm) can be obtained. The corresponding pair Pi0 ( equation M61, tci0′) is shown as the blue asterisk in figure 5. tci 0′ equals the Δtci of Pi. After the calibration phantom error is corrected, the computed microcalcification thickness is 245 μm, which is very close to the true thickness of 250 μm. Using this method, we can correct the calibration phantom error Δtc for all pixel pairs.

Figure 5
Demonstration of the correction method for calibration phantom error Δtc.

6. Conclusion

Dual-energy digital mammography can be used to suppress the contrast between adipose and glandular tissues to improve the detection of microcalcifications. It is a prospective technique to reduce the high false-negative rate of microcalcification detection in normal mammography. In this paper, the formulae of calibration phantom errors, which are caused by the differences of linear attenuation coefficients between calibration phantoms and breast tissues, were derived based on the imaging physical model. Based on the publicly available composition data of breast tissues and phantom materials, we calculated the calibration phantom errors and compared them with other types of errors in dual-energy digital mammography. Our results show that the attenuation differences between breast tissues and calibration phantoms can lead to significantly large calibration phantom errors in the order of tens to hundreds of percent, even if CIRS phantoms are adopted. A calibration phantom error is the major source of error in dual-energy digital mammography and varies for different breast compositions of different individuals.

After deeply investigating the physical meaning of calibration phantom errors, we found that calibration phantom errors are constant for a fixed glandular ratio. With this property and using all the pixels in the images, we suggested a correction method for the calibration phantom error of the microcalcification thickness. The results show that the large calibration phantom error of the microcalcification thickness can be corrected.

Acknowledgments

The project is partially supported by the National Science Fund of China (no 60472004 and no 60551003), the fund of the Ministry of Education of China (no 106143 and no NCET-05-0828), the fund of USA National Institutes of Health (EB007288) and the fund of ICRG of Hong Kong Polytechnic University (YG-79). The authors thank Ms Lena Ye for editorial refinements.

References

  • Alvarez RE, Macovski A. Energy-selective reconstruction in x-ray computerized tomography. Phys Med Biol. 1976;21:733–44. [PubMed]
  • Berger MJ, Hubbell JH, Seltzer SM, Chang J, Coursey JS, Sukumar R, Zucker DS. XCOM: Photon Cross Section Database (version 1.3) Gaithersburg, MD: National Institute of Standards and Technology; 2005. available at: http://physics.nist.gov/xcom.
  • Bliznakova K, Kolitsi Z, Pallikarakis N. Dual-energy mammography: simulation studies. Phys Med Biol. 2006;51:4497–515. [PubMed]
  • Boone JM, Shaber GS, Tecotzky M. Dual-energy mammography: a detector analysis. Med Phys. 1990;17:665–75. [PubMed]
  • Brandan ME, Ramirez RV. Evaluation of dual-energy subtraction of digital mammography images under conditions found in a commercial unit. Phys Med Biol. 2006;51:2307–20. [PubMed]
  • Brettle DS, Cowen AR. Dual-energy digital mammography using stimulated phosphor computed radiography. Phys Med Biol. 1994;39:1989–2004. [PubMed]
  • Brody WR, Butt G, Hall A, Macovski A. A method for selective tissue and bone visualization using dual energy scanned projection radiography. Med Phys. 1981;8:353–7. [PubMed]
  • Byng JW, Mainprize JG, Yaffe MJ. X-ray characterization of breast phantom materials. Phys Med Biol. 1998;43:1367–77. [PubMed]
  • Cardinal HN, Fenster A. An accurate method for direct dual-energy calibration and decomposition. Med Phys. 1990;17:327–41. [PubMed]
  • Chakraborty DP, Barnes GT. An energy sensitive cassette for dual-energy mammography. Med Phys. 1989;16:7–13. [PubMed]
  • Fandos-Morera A, Prats-Esteve M, Tura-Soteras JM, Traveria-Cros A. Breast tumors: composition of microcalcifications. Radiology. 1988;169:325–7. [PubMed]
  • Fewell TR, Shuping RE. Handbook of Mammographic X-ray Spectra. Rockville, MD: Bureau of Radiological Health; 1978. pp. 25–70.
  • Gingold EL, Hasegawa BH. Systematic bias in basis material decomposition applied to quantitative dual-energy x-ray imaging. Med Phys. 1992;19:25–33. [PubMed]
  • Hammerstein GR, Miller DW, White DR, Masterson ME, Woodward HQ, Laughlin JS. Absorbed radiation dose in mammography. Radiology. 1979;130:485–91. [PubMed]
  • Johns PC, Drost DJ, Yaffe MJ, Fenster A. Dual-energy mammography: initial experimental results. Med Phys. 1985;12:297–304. [PubMed]
  • Johns PC, Yaffe MJ. Theoretical optimization of dual-energy x-ray imaging with application to mammography. Med Phys. 1985;12:289–96. [PubMed]
  • Kappadath SC, Shaw CC. Dual-energy digital mammography: calibration and inverse-mapping techniques to estimate calcification thickness and glandular-tissue ratio. Med Phys. 2003;30:1110–7. [PubMed]
  • Kappadath SC, Shaw CC. Quantitative evaluation of dual-energy digital mammography for calcification imaging. Phys Med Biol. 2004;49:2563–76. [PubMed]
  • Kappadath SC, Shaw CC. Dual-energy digital mammography for calcification imaging: scatter and nonuniformity corrections. Med Phys. 2005;32:3395–408. [PubMed]
  • Kappadath SC, Shaw C, Lai CJ, Liu X, Whitman G. Dual-energy digital mammography for calcification imaging: theory and implementation. Proc SPIE. 2004;5368:751–60.
  • Kappadath SC, Shaw CC, Lai CJ, Liu X, Whitman GJ, Yang WT. Dual-energy digital mammography for calcification imaging: improvement by post-image processing. Proc SPIE. 2005;5745:1342–50.
  • Lehmann LA, Alvarez RE, Macovski A, Brody WR. Generalized image combinations in dual KVP digital radiography. Med Phys. 1981;8:659–67. [PubMed]
  • Lemacks MR, Kappadath SC, Shaw CC, Liu X, Whitman GJ. A dual-energy technique for microcalcification imaging in digital mammography: a signal-to-noise analysis. Med Phys. 2002;29:1739–51. [PubMed]
  • Mou XQ, Chen X. Error analysis of calibration materials on dual-energy mammography. Proc. 10th Int. Conf. on Medical Image Computing and Computer Assisted (Brisbane, Australia) LNCS; 2007. pp. 596–603. [PubMed]
  • Soo MS, Rosen EL, Xia JQ, Ghate S, Baker JA. Computer-aided detection of amorphous calcifications. Am J Roentgenol. 2005;184:887–92. [PubMed]
  • Taibi A, Fabbri S, Baldelli P, di Maggio C, Gennaro G, Marziani M, Tuffanelli A, Gambaccini M. Dual-energy imaging in full-field digital mammography: a phantom study. Phys Med Biol. 2003;48:1945–56. [PubMed]
  • White DR, Martin RJ, Darlinson R. Epoxy resin based tissue substitutes. Br J Radiol. 1977;50:814–21. [PubMed]